So we know republicans are going to pivot to defend ephebophilia. The question then becomes what rhetorical techniques work to make this look stupid.
I like to play dumb. Make them pronounce it and explain the definition. Finally, sigh and say; "there sure is a lot of nuance to fucking kids"
I like to play dumb. Make them pronounce it and explain the definition. Finally, sigh and say; "there sure is a lot of nuance to fucking kids"
November 14, 2025 at 6:04 PM
So we know republicans are going to pivot to defend ephebophilia. The question then becomes what rhetorical techniques work to make this look stupid.
I like to play dumb. Make them pronounce it and explain the definition. Finally, sigh and say; "there sure is a lot of nuance to fucking kids"
I like to play dumb. Make them pronounce it and explain the definition. Finally, sigh and say; "there sure is a lot of nuance to fucking kids"
Good news libtards! We're about to have our first openly bi president
November 14, 2025 at 5:31 PM
Good news libtards! We're about to have our first openly bi president
ICE is a front for domestic surveillance technology.
September 21, 2025 at 6:09 PM
ICE is a front for domestic surveillance technology.
July 24, 2025 at 5:28 PM
Just a thought; what about a #cryptocurrency project where every dollar made from tokens was spent directly on compute for training #opensource #LLMs. Token holders get unlimited API access? WIP
July 23, 2025 at 5:23 PM
Just a thought; what about a #cryptocurrency project where every dollar made from tokens was spent directly on compute for training #opensource #LLMs. Token holders get unlimited API access? WIP
New study: Quantization Bounds in LoRA Fine-tuning
• Theory: excess risk = √r/√N + r·2⁻²ᵇ σ² → derive optimal bits b* ≈ ½ log₂(rN)
• DialoGPT tests:
8-bit ≈ FP16 for ranks ≤16
4-bit OK to rank 8, breaks after
grad-var ∝ r·2⁻²ᵇ
RoT: higher LoRA rank ⇒ use higher precision. #LoRA #LLM #AI
• Theory: excess risk = √r/√N + r·2⁻²ᵇ σ² → derive optimal bits b* ≈ ½ log₂(rN)
• DialoGPT tests:
8-bit ≈ FP16 for ranks ≤16
4-bit OK to rank 8, breaks after
grad-var ∝ r·2⁻²ᵇ
RoT: higher LoRA rank ⇒ use higher precision. #LoRA #LLM #AI
github.com
July 9, 2025 at 6:27 PM
Just dropped: Adaptive LoRA
We tested non-uniform LoRA rank allocation (linear decay, attention-heavy, empirical) on DialoGPT.
Findings:
• Fixed-rank LoRA (r=16) still wins (PPL: 134.0)
• Linear decay is stable but weaker (PPL: 165.1)
• Complex strategies = unstable
🎯 Adaptive LoRA ≠ magic—yet.
We tested non-uniform LoRA rank allocation (linear decay, attention-heavy, empirical) on DialoGPT.
Findings:
• Fixed-rank LoRA (r=16) still wins (PPL: 134.0)
• Linear decay is stable but weaker (PPL: 165.1)
• Complex strategies = unstable
🎯 Adaptive LoRA ≠ magic—yet.
July 8, 2025 at 9:55 AM
Just dropped: Adaptive LoRA
We tested non-uniform LoRA rank allocation (linear decay, attention-heavy, empirical) on DialoGPT.
Findings:
• Fixed-rank LoRA (r=16) still wins (PPL: 134.0)
• Linear decay is stable but weaker (PPL: 165.1)
• Complex strategies = unstable
🎯 Adaptive LoRA ≠ magic—yet.
We tested non-uniform LoRA rank allocation (linear decay, attention-heavy, empirical) on DialoGPT.
Findings:
• Fixed-rank LoRA (r=16) still wins (PPL: 134.0)
• Linear decay is stable but weaker (PPL: 165.1)
• Complex strategies = unstable
🎯 Adaptive LoRA ≠ magic—yet.
Just finished some research on selective placement of low ranking adapters in inference models. Check it out and lmk if youre an endorser on arxiv! #machinelearning #ai #cs github.com/CatsMeow492/...
github.com
June 8, 2025 at 10:03 PM
Just finished some research on selective placement of low ranking adapters in inference models. Check it out and lmk if youre an endorser on arxiv! #machinelearning #ai #cs github.com/CatsMeow492/...
We a pro 2a party again yet?
June 8, 2025 at 9:59 PM
We a pro 2a party again yet?
I feel like I’m on the precipice of a very interesting question; in CS we’ve largely mapped out how language and and vision work. But what about economic utility/value? At my startup we’re trying to map out these relationships via the same methodology as language and it’s falling short.
March 16, 2025 at 6:50 PM
I feel like I’m on the precipice of a very interesting question; in CS we’ve largely mapped out how language and and vision work. But what about economic utility/value? At my startup we’re trying to map out these relationships via the same methodology as language and it’s falling short.
a blind god sifting stardust for constellations it cannot see.
February 22, 2025 at 11:47 PM
a blind god sifting stardust for constellations it cannot see.
The messiah effect is a consequence of monotheism, in which the population believes that social change comes all at once as a result of a single historical figure. Change is more like water gradually eroding rock over centuries.
February 4, 2025 at 4:52 PM
The messiah effect is a consequence of monotheism, in which the population believes that social change comes all at once as a result of a single historical figure. Change is more like water gradually eroding rock over centuries.
It might be totally immoral but FYI; if you call a conservative gay or any variation of gay they respond like a possum or a ragdoll cat. I’m not sure why. Evolution or selective breeding is my guess.
January 25, 2025 at 11:35 AM
It might be totally immoral but FYI; if you call a conservative gay or any variation of gay they respond like a possum or a ragdoll cat. I’m not sure why. Evolution or selective breeding is my guess.
My god Elon…
January 16, 2025 at 7:23 PM
My god Elon…
Reposted
On some level I know that China having our data is more dangerous than an American company because of the lack of laws and governance.
But with GOP leadership, I don’t think that’s true anymore. Just look at what happened with our DNA with Ancestry…
But with GOP leadership, I don’t think that’s true anymore. Just look at what happened with our DNA with Ancestry…
January 15, 2025 at 10:20 PM
On some level I know that China having our data is more dangerous than an American company because of the lack of laws and governance.
But with GOP leadership, I don’t think that’s true anymore. Just look at what happened with our DNA with Ancestry…
But with GOP leadership, I don’t think that’s true anymore. Just look at what happened with our DNA with Ancestry…
Anybody got some of those contraband memes. You know… the good shit from TikTok?
January 13, 2025 at 5:26 PM
Anybody got some of those contraband memes. You know… the good shit from TikTok?
I had my suspicions Elon Musk was an idiot but I’ve never been more certain of anything after seeing the Path of Exile 2 stream. Could you pick a sweatier game with a more unbathed fan base to try and bullshit?
January 11, 2025 at 6:54 PM
I had my suspicions Elon Musk was an idiot but I’ve never been more certain of anything after seeing the Path of Exile 2 stream. Could you pick a sweatier game with a more unbathed fan base to try and bullshit?
Absolutely crazy to me that conservatives are the majority in every branch of the federal government and conservative voters are able to maintain this level of cope.
“Be prepared to fight to get the Dems out of the fed government.”
“Be prepared to fight to get the Dems out of the fed government.”
January 4, 2025 at 5:21 PM
Absolutely crazy to me that conservatives are the majority in every branch of the federal government and conservative voters are able to maintain this level of cope.
“Be prepared to fight to get the Dems out of the fed government.”
“Be prepared to fight to get the Dems out of the fed government.”
Having multiple family members that served in special forces I can tell you with one hundred percent certainty that if a 15 year veteran of the Green Berets wanted to kill people in mass - they would be successful.
apnews.com/article/trum...
apnews.com/article/trum...
Soldier shot self in head before Cybertruck exploded outside Trump's Las Vegas hotel, officials say
Matthew Livelsberger was a member of the Army’s elite Green Berets, according to an Army statement.
apnews.com
January 2, 2025 at 8:55 PM
Having multiple family members that served in special forces I can tell you with one hundred percent certainty that if a 15 year veteran of the Green Berets wanted to kill people in mass - they would be successful.
apnews.com/article/trum...
apnews.com/article/trum...
The best strategy to my knowledge at this point in the cycle of fascism is unfortunately likely silence. Be a facilitator - if mass deportations happen help people find their way home and give comfort. Economic and political realities usually drive the rest of what needs to happen.
January 1, 2025 at 5:20 AM
The best strategy to my knowledge at this point in the cycle of fascism is unfortunately likely silence. Be a facilitator - if mass deportations happen help people find their way home and give comfort. Economic and political realities usually drive the rest of what needs to happen.
Useful idiots are being purged and shadowbanned from Twitter post election.
December 27, 2024 at 4:13 PM
Useful idiots are being purged and shadowbanned from Twitter post election.